CN106846263A - The image defogging method being immunized based on fusion passage and to sky - Google Patents

Abstract

The image defogging method being immunized based on fusion passage and to sky, it is related to digital image processing techniques field, there is sky areas color distortion on image after solving conventional images defogging method defogging, there is halation phenomenon and blocking effect, defog effect difference and existing method is caused to there are problems that identification process troublesome calculation amount, minimum value, maximum value filtering are carried out to foggy image, rough estimate atmosphere light and introducing modifying factor, the respectively rough estimate of the transmitance of the transmitance and bright passage of dark；Fine optimization is carried out using guiding filtering respectively, the rough estimate made a rough estimate of whether more than bright passage transmitance of dark transmitance is judged, if, dark primary channel image is sky areas, if it is not, then dark primary channel image is not sky areas, obtains accurate fusion transmitance and estimate；And the pixel average is accurately estimated and fusion transmitance as the self adaptation to atmosphere light, defogging sharpening image is obtained using atmospheric scattering imaging model.

Description

The image defogging method being immunized based on fusion passage and to sky

Technical field

The present invention relates to digital image processing techniques field, and in particular to it is a kind of based on fusion passage and to sky be immunized Image defogging method.

Background technology

In recent years, haze weather is relatively conventional, due in air the particle such as a large amount of particulates, small water droplet to light Absorption and scattering process cause outdoor image quality seriously to be degenerated, color of image distortion is partially greyish white, and image blurring details is unclear Clear, contrast declines.Visual effect is not only influenceed, the performance of the effectiveness of outdoor imaging system is also directly limit, later stage figure is given As treatment and analysis bring difficulty.Therefore, sharpening treatment is carried out to Misty Image very necessary.

At present, image defogging technology has become study hotspot, have to single image to the fog method image enchancing method and Two kinds of image recovery method.Image enchancing method is mainly by the means such as image procossing, such as histogram equalization, frequency domain filter Ripple, Retinex etc., by weakening or removing some of some unwanted information, prominent image information, to a certain extent The contrast of image can be improved, improves visual effect, but do not started with from the blur degradation mechanism of Misty Image, be not reality Defogging in matter.

Image recovery method defogging is from atmospheric scattering theory, it is considered to image degradation reason, sets up greasy weather imaging Scattering model, realizes that scene is restored, and ideal defog effect can be obtained, wherein with the defogging method of dark primary priori as generation Table.Dark primary priori is a kind of statistical law to fog free images, and rough estimate first goes out Medium Propagation function, is then scratched using soft Primitive is managed or wave filter carries out fine optimization to transmitance image, and the algorithm obtains admirable defog effect.

But there is the sky areas base of large area in the case where the brightness of scene objects is similar to atmosphere light, in image Originally the point of pixel value very little is can not find, now dark primary assumes failure.Typically can all there is certain area for outdoor image Sky image, however, either algorithm for image enhancement or Image Restoration Algorithm above, these algorithms have one common to lack Point, can exactly cause sky areas color distortion after defogging, halation phenomenon and blocking effect occur, have a strong impact on defog effect.Phase The innovatory algorithm answered needs to extract the features such as sky gradient information, carries out sky identification, so as to enter to sky and non-sky areas Row segmentation, then takes different transmitance methods of estimation to carry out defogging treatment, and identification process is cumbersome, computationally intensive.

The content of the invention

There is halation to there is sky areas color distortion on image after solving conventional images defogging method defogging in the present invention Phenomenon and blocking effect, cause defog effect difference and existing method to there are problems that identification process troublesome calculation amount, there is provided one Plant the image defogging method being immunized based on fusion passage and to sky.

The image defogging method being immunized based on fusion passage and to sky, the method is realized by following steps：

Step one, carry out mini-value filtering and maximum value filtering respectively to original foggy image I (x), obtain dark former respectively Chrominance channel image and bright primary channel image；

Step 2, the pixel for taking brightness value highest 0.1% in dark primary channel image, in original foggy image I (x) The point of respective pixel value is found, and takes the pixel value of maximum as the rough estimate to atmosphere light A；

Step 3, introduce modifying factor ω, and according to step 2 obtain atmosphere light A rough estimate, obtain dark The rough estimate of transmitanceWith the rough estimate of bright passage transmitanceIt is expressed as with following formula：

In formula, c is a Color Channel in R, G, B triple channel, takes the modifying factor ω of dark primary channel image_D= 0.8, the modifying factor of bright primary channel image

Step 4, the rough estimate that dark transmitance is obtained according to step 3Rough with bright passage transmitance is estimated MeterUsing guiding filtering respectively to the rough estimate of dark transmitanceWith the rough estimate of bright passage transmitanceFine optimization is carried out, the fine estimation of dark transmitance is respectively obtainedWith the fine estimation of bright passage transmitance

Step 5, the fine estimation for judging dark transmitanceWhether the fine estimation of bright passage transmitance is more thanIf it is, performing step 6；If it is not, then dark primary channel image is not sky areas, fusion transmitance is estimated

Step 6, the dark primary channel image are sky areas, and fusion transmitance is estimatedStatistics symbol The pixel value of the corresponding foggy image part in sky areas of bright primary colors priori theoretical is closed, and using the pixel average as to big The self adaptation of gas light A is accurately estimated；

Step 7, according to the fusion transmitance comprehensively obtained in step 5, step 6The air obtained in step 6 The self adaptation of light A is accurately estimated, using atmospheric scattering imaging model acquisition defogging sharpening image.

Beneficial effects of the present invention：The present invention is based on atmospherical scattering model, in the theoretical foundation of dark primary priori, from Secretly, bright binary channels is started with and is analyzed, and the situation of dark primary priori failure is made up using bright primary colors priori, to sky portion image There is the immunity of self adaptation, and accurately ART network can be carried out to atmosphere light A, effectively increase dark primary elder generation checking method Universality and robustness.

Brief description of the drawings

Fig. 1 is outdoor fog free images dark primary image of the invention and bright primary colour image；Wherein, Fig. 1 a are outdoor fogless original Image, Fig. 1 b are dark primary image, and Fig. 1 c are bright primary colour image；

Fig. 2 is a kind of image defogging method flow chart being immunized based on fusion passage and to sky of the invention；

The contrast effect estimated transmitance image according to dark, bright passage and fusion passage is respectively in Fig. 3 Figure；Wherein, Fig. 3 a are artwork, and Fig. 3 b finely estimate transmitance design sketch for dark primary passage, and 3c is that bright primary channel estimates essence Thin transmitance design sketch, Fig. 3 d estimate transmitance design sketch for fusion passage；

Fig. 4 is using the image defogging method and existing Nogata being immunized based on fusion passage and to sky of the present invention Figure equalization algorithm, dark channel prior algorithm are to the defog effect contrast effect figure with small area sky mist figure；Wherein, Fig. 4 a It is artwork, Fig. 4 b are to the defog effect figure with small area sky mist figure using algorithm of histogram equalization；Fig. 4 c are use To the defog effect figure with small area sky mist figure, Fig. 4 d are based on fusion using of the present invention to dark channel prior algorithm Passage and to sky be immunized image defogging method to the defog effect figure with small area sky mist figure；

Fig. 5 is using the image defogging method and existing Nogata being immunized based on fusion passage and to sky of the present invention Figure equalization algorithm, dark channel prior algorithm are to the defog effect comparison diagram with large area sky mist figure；Wherein, Fig. 5 a are original Figure, Fig. 5 b are to the defog effect figure with large area sky mist figure using algorithm of histogram equalization；Fig. 5 c are using helping secretly To the defog effect figure with large area sky mist figure, Fig. 5 d are based on fusion passage using of the present invention to road elder generation checking method And the image defogging method being immunized to sky is to the defog effect figure with large area sky mist figure；

Fig. 6 is based on this algorithm with algorithm of histogram equalization, dark channel prior algorithm to large area and sky and thing The folded Misty Image defog effect figure of body weight；Wherein, Fig. 6 a are artwork, and Fig. 6 b are to big using algorithm of histogram equalization The Misty Image defog effect figure of area and sky and overlapped object；Fig. 6 c are to large area using dark channel prior algorithm And the Misty Image defog effect figure of sky and overlapped object, Fig. 6 d are based on fusion passage and to day using of the present invention The immune image defogging method of sky is to the Misty Image defog effect figure with large area and sky and overlapped object.

Specific embodiment

Specific embodiment one, with reference to Fig. 1 to Fig. 6 illustrate present embodiment, based on fusion passage and to sky be immunized Image defogging method, the method is realized by following steps：

Step one, carry out minimum value, maximum value filtering respectively to original foggy image I (x), dark primary, bright is asked for respectively The image of primary channel.

Step 2,0.1% pixel before dark channel image sorts according to pixel value size is taken, and in original foggy image Respective pixel value point is found, the pixel value of maximum is taken as the rough estimate to atmosphere light A.

Step 3, the rough estimate by step 2 to atmosphere light A, dark is respectively obtained according to formula (5), formula (10) TransmitanceWith the transmitance of bright passageRough estimate；

Its detailed process is：

In computer vision field, atmospheric scattering imaging model is used widely, is shown below：

I (x)=J (x) t (x)+A (1-t (x)) (1)

Wherein, J (x) is the fog free images to be recovered, and t (x) is transmitance, and c refers to that a color in R, G, B triple channel is led to Road.Formula (1) is slightly processed, following formula is deformed into：

By dark primary priori theoretical, assume initially that transmitance t (x) is constant in each window, is defined asA It is constant, minimum operation twice is asked to formula (2) both sides, obtains following formula：

It can be seen from dark channel prior：

In order to allow people to feel the presence of the depth of field, it is necessary to defogging when targetedly retain a part remote scape of covering The mist of thing, introduces a modifying factor ω between [0,1], finally can obtain the rough estimate of dark transmitance

Wherein, the modifying factor ω of dark primary channel image_DValue is：ω_D=0.8.

Found for fog free images statistics in Outdoor Scene, in any local fritter of most open air fog free images, also deposited In some pixels, the intensity level of their some or several Color Channels is very high, or even close to 255 saturation values, herein I Be referred to as bright primary colors.

Through analysis, outdoor fogless topography meets dark primary and bright primary colors, but Ye You topographies can not be simultaneously simultaneously Meet both prioris, as shown in Figure 1.Obviously, sky areas topography only meets bright primary colors priori, and is unsatisfactory for dark Primary colors priori, then mistake will occurs using dark, causes the distortion of sky portion color of image, halation phenomenon and block occurs Effect.

We copy dark to push over process, and bright passage is pushed over as follows：Maximum twice is asked for formula (2) both sides Computing, obtains formula：

From bright channel prior：

By (6) Shi Ke get：

Arrangement can be obtained：

Copy dark theoretical, image is normalized, and add modifying factor, formula (10) is obtained, using this Individual formula can be obtained using the rough estimate of bright passage transmitance

Wherein, the modifying factor ω of bright primary channel image_LValue is：

From formula (5)：It isSubtraction function, illustrates to work asValue is bigger, i.e. transmitance figure is got over It is bright,With regard to smaller, then it is bigger that the topography meets dark channel prior rule probability, conversely, transmitance figure Darker, it is smaller that the topography meets dark channel prior rule probability.

From formula (10)：It isIncreasing function, illustrate work asValue is bigger, i.e. transmitance Figure is brighter,It is bigger, then it is bigger that the topography meets bright channel prior rule probability, conversely, thoroughly Cross rate figure darker, it is smaller that the topography meets dark channel prior rule probability.

Step 4, the rough estimate by step 3 to the transmitance of dark, bright passage Using guiding filtering point It is not right Fine optimization estimation is carried out, the fine estimation of dark transmitance is respectively obtainedIt is saturating with bright passage Cross the fine estimation of rateArtwork to the Misty Image in Fig. 3 a carries out fine transmitance estimation, respectively obtains dark primary Passage finely estimates transmitance figure such as Fig. 3 b, and bright primary channel finely estimates transmitance figure such as Fig. 3 c, Fig. 3 d to melt in the present invention Close passage and estimate transmitance figure.

Step 5, judgement Relation come obtain it is more accurate fusion transmitance estimation figureChoose The fine estimation of dark transmitanceWith the fine estimation of bright passage transmitanceMiddle relative larger value is (such as formula (11) It is shown), obtain final transmitance and estimateSo, can cause that dark primary and bright primary colors theory mutually make up to obtain Take more accurate transmitance estimation figure

IfExplanation is sky portion, then improved fusion transmitanceConversely, then It is not sky portion, improved fusion transmitanceFinally give improved fusion transmitance figureSuch as Fig. 3 d It is shown.

Step 6, take step 5,The sky portion of dark primary priori theoretical is not met to that there should be mist figure As the pixel value of part, average is counted and asked for, accurately estimated as the self adaptation to atmosphere light A.

Step 7：Take fusion transmitance figure in step 5, step 6The fine ART network value of atmosphere light A, root Defogging sharpening image is obtained according to atmospheric scattering imaging model.

In present embodiment, three kinds of foggy images that there is sky, a kind of greasy weather figure for being to exist small area sky are chosen Picture, as shown in Figure 4；Second is the Misty Image with large area sky, as shown in Figure 5；The third is with large area day The Misty Image that empty and sky overlaps with object, as shown in Figure 6；

Choose typical defogging algorithm in typicalness algorithm-algorithm of histogram equalization, the image restoration of image enhaucament-dark Channel prior algorithm, and a kind of image defogging algorithm being immunized based on fusion passage and to sky of the invention is to the image in greasy weather Sharpening treatment is carried out, and defog effect is contrasted, result is referring to Fig. 4-Fig. 6.

Contrast defog effect shows：In Misty Image, whether no matter there is sky size, histogram equalization is calculated Can all there is color distortion, halation phenomenon and blocking effect in method, dark channel prior algorithm, ginseng to the result of sky portion image 4b, 4c, 5b, 5c, 6b and the 6c seen in Fig. 4-Fig. 6, have a strong impact on the defog effect of image.

Using the figure obtained based on fusion passage described in present embodiment and to the image defogging algorithm process that sky is immunized Picture, no matter sky portion image area size, effectively prevent distortion of the algorithm above to Misty Image sky portion, obtain whole The harmonious clearly image of body, illustrates the immunity of the sky portion that inventive algorithm is processed Misty Image.

Claims (3)

1. the image defogging method being immunized based on fusion passage and to sky, it is characterized in that, the method is realized by following steps：

Step one, carry out mini-value filtering and maximum value filtering respectively to original foggy image I (x), dark primary is obtained respectively and is led to Road image and bright primary channel image；

Step 2, the pixel for taking brightness value highest 0.1% in dark primary channel image, find in original foggy image I (x) The point of respective pixel value, and the pixel value of maximum is taken as the rough estimate to atmosphere light A；

Step 3, introduce modifying factor ω, and according to step 2 obtain atmosphere light A rough estimate, obtain dark pass through The rough estimate of rateWith the rough estimate of bright passage transmitanceIt is expressed as with following formula：

${\tilde{t}}_D\left(x\right)=1-{\mathrm{\ω}}_D\underset{y\∈\mathrm{\Ω}\left(x\right)}{\min }\left(\underset{c}{\min }\frac{I^c\left(x\right)}{A^c}\right)$

${\tilde{t}}_L\left(x\right)=1-{\mathrm{\ω}}_L\frac{1-\underset{y\∈\mathrm{\Ω}\left(x\right)}{max}\left(\underset{c}{max}{I}^c\right(x\left)\right)}{1-A^c}$

In formula, c is a Color Channel in R, G, B triple channel, takes the modifying factor ω of dark primary channel image_D=0.8, it is bright The modifying factor of primary channel image

Step 4, the rough estimate that dark transmitance is obtained according to step 3With the rough estimate of bright passage transmitanceUsing guiding filtering respectively to the rough estimate of dark transmitanceWith the rough estimate of bright passage transmitanceFine optimization is carried out, the fine estimation of dark transmitance is respectively obtainedWith the fine estimation of bright passage transmitance

Step 5, the fine estimation for judging dark transmitanceWhether the fine estimation of bright passage transmitance is more than If it is, performing step 6；If it is not, then dark primary channel image is not sky areas, fusion transmitance is estimated

Step 6, the dark primary channel image are sky areas, and fusion transmitance is estimatedStatistics meets bright original The pixel value of the corresponding foggy image part in sky areas of color priori theoretical, and using the pixel average as to atmosphere light A Self adaptation accurately estimate；

Step 7, the fusion transmitance comprehensively obtained according to step 5, step 6The atmosphere light A's obtained in step 6 Self adaptation is accurately estimated, using atmospheric scattering imaging model acquisition defogging sharpening image.

2. it is according to claim 1 based on fusion passage and to the immune image defogging method of sky, it is characterised in that institute State the rough estimate of bright passage transmitanceThe detailed process of acquisition is：

Atmospheric scattering imaging model both sides are asked for maximum operation twice, following formula is obtained：

$\underset{y\∈\mathrm{\Ω}\left(x\right)}{max}\left(\underset{c}{max}\frac{I^c\left(x\right)}{A^c}\right)={\tilde{t}}_L\left(x\right)\underset{y\∈\mathrm{\Ω}\left(x\right)}{max}\left(\underset{c}{max}\frac{J^c\left(x\right)}{A^c}\right)+1-{\tilde{t}}_L\left(x\right)$

It can be seen from bright primary colors priori：

$\underset{y\∈\mathrm{\Ω}\left(x\right)}{max}\left(\underset{c}{max}\frac{J^c\left(x\right)}{A^c}\right)\≤\frac{255}{A^c}$

Arrangement can be obtained：

${\tilde{t}}_L\left(x\right)\≥\frac{{\underset{y\∈\mathrm{\Ω}\left(x\right)}{max}}_{}\left({\max }_c{I}^c\right(x\left)\right)-A^c}{255-A^c}$

Image is normalized, and bright primary channel image modifying factor ω_L, obtain the rough of bright passage transmitance EstimateIt is expressed as with following formula：

${\tilde{t}}_L\left(x\right)=1-{\mathrm{\ω}}_L\frac{1-\underset{y\∈\mathrm{\Ω}\left(x\right)}{max}\left(\underset{c}{max}{I}^c\right(x\left)\right)}{1-A^c}.$

3. it is according to claim 1 based on fusion passage and to the immune image defogging method of sky, it is characterised in that right Merge the estimation of transmitanceIts detailed process is：

Choose the fine estimation of dark transmitanceWith the fine estimation of bright passage transmitanceMiddle relative larger value is made For fusion transmissivity is estimatedIt is expressed as with following formula：

${\tilde{t}}_E\left(x\right)=max\left({\tilde{T}}_D\right(x),{\tilde{T}}_L(x\left)\right).$